AI is a helpful tool when used properly, but like all tools it has limitations. It is, for example, not recommended to use AI to write a research paper. That’s not to say that AI can’t be used for research at all — AI is, for instance, pretty nifty at summarizing, which happens to be a very helpful tool for one ubiquitous task in research: a literature review.

Literature reviews are important, and are included in nearly every published paper, for several reasons. They set the stage for a paper’s main idea, they place the research into the context of the rest of the field for readers, and they help the author to find and understand the work that’s come before them (in many cases, work they will directly build upon).

Even so, the process of starting a literature review is rarely anyone’s favorite task, as it can be a rather tedious process of finding relevant papers amidst a sea of scholarly work. And literature reviews themselves are seldom the most exciting part of a new paper.

In days of old, scholars had to travel to libraries, pore over the titles of dozens of tomes, and seek out relevant chapters of books they’d need to read. The Internet made it much easier and faster to find relevant work, particularly with the help of resources like Google Scholar, and particularly by looking for papers with high numbers of citations within their topic area. This is still a great way to search for relevant work, and should not be neglected. Especially for researchers without years of experience, this can be a time-consuming and challenging process, and it’s easy to miss relevant work.

How to use AI to start your literature review

Asking AI about the research landscape on a certain topic can be a very helpful way to get started, with the results offering an efficient first step to begin a literature review, especially for younger researchers or economists shifting to a new topic. That is: you can simply ask AI to summarize the research field for you. Careful wording, as always, will help to improve your output.

The aim of using this tool is to help you jump-start your understanding of existing research in your topic area, not to actually write anything for you. Rather than spend hours searching through dozens of journals and papers for relevant topics and poring over them until you get an idea of what’s mainstream and seminal, AI can provide a sufficient overview of the research in your topic, and point out the seminal works that make excellent starting points when building your topic knowledge.

It’s important that you read the AI output thoroughly and use it to build a reading plan for yourself. You’ll need to be familiar with the most important papers, and use both the AI’s suggestions and the actual papers’ lists of citations to start branching out into new directions. It’s especially important to look for recent papers that are on the cutting-edge of research in the field and see how these relate to older, “classic” works. Otherwise, you might end up researching something that’s already been done!

Of course, hallucination is an ever-present issue with AI, but for this case, assuming you follow the advice and find and read each paper, it won’t be an issue at all — if AI hallucinates a paper, you will not be able to find it.

And if your AI tool hallucinates in this way, it can even help you find research gaps in the literature — if AI invented a paper to try and answer one of your questions, it suggests that question may not be sufficiently answered by existing research. That can be exciting for a new researcher, but remember to still search for anything relevant before proclaiming your research idea as the first of its kind.

AI can help you identify originality, as well — simply asking it if your research question has been studied before is a nice first step. Still, it’s critical to remember that AI is not infallible and it cannot substitute for reading and learning. A dedicated researcher will need to digest the relevant work and understand it to such a degree that they can answer live questions from more senior researchers and lecturers during conferences, presentations, and their thesis defense. AI cannot help you present your work.

An example query

Here’s a query that I wrote while exploring topic areas during my economics PhD, using the free version of ChatGPT (GPT-5.6 Luna):

“I am beginning to research housing and how it relates to wealth inequality in economics. I am a new researcher without much experience in the field, and I need to know what the most important and relevant papers are. Please give me a list of the most relevant papers about the intersection of housing and wealth inequality, preferably from an applied micro standpoint (but please mention any highly-cited and relevant macro-focused papers related to this topic in a separate section). Please include a quick, two-sentence summary of each paper.”

The AI’s response to this prompt immediately gave me great places to dig deeper. First, it pointed out that housing and wealth inequality are very broad topics and that it’d be more productive to focus on something more specific. The suggestions it made, based on what research had already been done, were: “homeownership and wealth accumulation”, “house-price appreciation and wealth inequality”, “intergenerational transmission”, “mortgage/credit constraints”, and “housing supply and location”.

Then, the AI recommended a whopping 20  papers and gave suggestions about which ones were best to start with. These included:
1. Causa, Woloszko, and Leite (2019): “Housing, Wealth Accumulation and Wealth Distribution: Evidence and Stylized Facts”. This is a data-focused paper that uncovers some present facts about wealth distributions, and is a good foundational read for someone interested in the topic for modern research purposes.
2. OECD (2021): “Measuring the Role of Housing in the Distribution of Wealth”. This paper is a chapter in the OECD’s “Brick by Brick” report. It is another paper that helps understand the foundations of the topic, this time by going deep into the relationship between homeownership and wealth inequality.
3. Sodini, Van Nieuwerburgh, Vestman, and von Lilienfeld-Toal (2023): “Identifying the Benefits from Homeownership: A Swedish Experiment”. This empirical paper examines the causal link between homeownership and wealth accumulation and consumption smoothing using a quasi-random natural experiment.
4. Campbell and Cocco (2007): “How Do House Prices Affect Consumption? Evidence from Micro Data”. Unsurprisingly, this empirical paper does exactly what it says: examines how changes in house prices affect consumption decisions.

Helpfully, all of the papers came with links to the source the AI found them in, allowing me to find them extremely quickly. This is something that AI models are doing increasingly, reducing the amount they hallucinate.

Still, while including links helps the models to cut down on full hallucinations, I’ve found that AI models may still provide papers that are relevant on first glance but don’t contain useful content related to the question on a deeper dive. This provides another reason to read the output carefully — which you should be doing anyway if you intend to use those papers as part of a literature review.

If an AI model provided a paper in an attempt to answer your query, but the paper turned out to not be incredibly relevant, this can suggest that the research might be lacking in that area and that your question ought to be studied further. This again highlights another helpful benefit from using AI to start a literature review: it can help you identify whether your research question is specific and new enough to dig into.

As I demonstrated, doing this can also help you learn the basics of a field before you approach colleagues, or start digging deep into specific papers…

The role of colleagues and superiors

If using AI in this way feels blasphemous, consider the following. Querying AI to overview your research field is akin to walking into a colleague’s office and asking them to give you an overview about the field themselves, but considerably faster and uses less of others’ valuable time.

You can and should still ask your colleagues and supervisors about your research ideas! But, in this case it may be better (and more considerate) to ask AI to give you an overview first. This will prevent you from needing to wait until schedules align to get started, and will allow you to kick-start a more involved — and informed — discussion about the topic, rather than have your colleagues simply list off relevant papers they’re aware of.

Being prepared before you meet with colleagues about a specific topic reflects well on you, and your ability.

Do not use AI to write your literature review!

Used in this way, AI is a fantastic supplemental tool that can quickly get you up to speed and help you direct your focus and efforts in the right direction. Research is difficult; anything that can save you time will help you to perfect the more original aspects of your work.

Even now, research is being done about the best ways to use AI in literature reviews. For example, this open access study in Springer recommends using AI for “screening and data extraction”. Feel free to dive deeper into their paper for best practices if you’d like to see more.

While hallucination isn’t an issue if you use AI in this way, it will be an issue if you try to have AI simply write your literature review. It can and will invent or misappropriate papers to fill gaps that exist, and make up links, authors or conclusions that don’t exist. This can cause serious problems for a researcher trying to submit a paper, and cast doubt on your ability and dedication.

And, as always, you owe it to yourself to do the work. It’ll make you a better version of the economist you want to be!

---

^{Header Image Credit: pressfoto on Magnific (free license)}